Dice score is not change

[JonyJiang123]

Recently, I got the problem when trained model with own data.

Dice score is unchanged and close to 0， but the same data is in another unet version(keras), test result is good.

INFO: Starting training: Epochs: 50 Batch size: 16 Learning rate: 0.0001 Training size: 232 Validation size: 25 Checkpoints: True Device: cuda Images scaling: 1.0 Mixed Precision: False

[Huster-Hq]

You can try changing the optimizer

[nqthai309]

have you solved it yet? help me

[VladDok]

@biel-jony, try to reduce learning rate and check result.

[leosmaia21]

Hello, did you manage to solve it?